davis.metadata.json

{
  "bias": {
    "caveat": "Shares compare where reports land against where exposure is. They cannot, on their own, separate 'more dangerous' from 'more reported': reporter pools skew by route choice, demographics, app access, and language. Treat over-represented segments as candidates for attention and scrutiny, not as confirmed rankings.",
    "over_represented": [
      {
        "exposure_share": 0.0159,
        "report_share": 0.1053,
        "segment_id": "seg-06"
      },
      {
        "exposure_share": 0.0212,
        "report_share": 0.1053,
        "segment_id": "seg-02"
      },
      {
        "exposure_share": 0.0212,
        "report_share": 0.1053,
        "segment_id": "seg-05"
      }
    ],
    "under_represented": [
      {
        "exposure_share": 0.0794,
        "report_share": 0.0,
        "segment_id": "seg-12"
      },
      {
        "exposure_share": 0.0794,
        "report_share": 0.0,
        "segment_id": "seg-09"
      },
      {
        "exposure_share": 0.4233,
        "report_share": 0.3509,
        "segment_id": "seg-03"
      }
    ]
  },
  "city": "Davis",
  "corridor_dataset": "davis.corridors.geojson",
  "corridor_maup_note": "Corridor rates are a second, coarser view over the same already-published block-level segments — a MAUP-aware complement, not a replacement. Both granularities are published side by side; a corridor's count and exposure are sums of segments that individually already cleared the k-anonymity floor and the Getis-Ord significance test, so corridors never surface a block that was withheld or non-significant on its own. Aggregation choice (the boundary you draw) can itself shift a rate — the Modifiable Areal Unit Problem — which is exactly why the block-level file remains the primary published unit and this file is always secondary to it.",
  "corridors_published": 1,
  "data_card": "docs/DATA-CARD.md",
  "dataset_note": "Synthetic demonstration data — not real reports. The hotspots are planted in the fixture, so the method finds what was put there.",
  "geojson_sha256": "45ae40f0f295a604a78976af8a35c981fc62c06b63c96a9820aed5f08f1790d2",
  "license": "Apache-2.0",
  "maup_rank_stability": {
    "basis": "greedy nearest-neighbour re-segmentation to a coarser partition",
    "coarse_units": 90,
    "fine_units": 180,
    "k": 5,
    "top_hotspot_coarse_rank": 1,
    "top_hotspot_segment": "seg-06",
    "top_hotspot_still_significant": true,
    "top_hotspot_survives": true,
    "topk_overlap": 0.8
  },
  "methods": {
    "confidence_z": 1.96,
    "dispersion": {
      "adjusted": false,
      "model": "quasi-Poisson (Pearson) on the rate/offset model",
      "phi": 11.1761
    },
    "exposure_floor": 0.0,
    "exposure_stale_days": 365.0,
    "fdr_alpha": 0.05,
    "getis_ord_band_m": 300.0,
    "getis_ord_neighbors": "street-network adjacency/distance",
    "getis_ord_node_snap_m": 5.0,
    "kde_bandwidth_m": 150.0,
    "min_publish_n": 3,
    "rate_definition": "pooled across all hazard types (union); per-type rates in rates_by_type where count >= small_n",
    "rate_per": 1000.0,
    "significance": "Getis-Ord Gi* on the exposure-normalized rate, Benjamini-Hochberg FDR",
    "small_n": 5
  },
  "privacy": "Aggregated to public street segments; segments with a non-zero report count below min_publish_n are withheld (k-anonymity). No per-report coordinate, time, reporter token, note, mode, or severity is published. Small-n hazard breakdowns are suppressed.",
  "report_intensity_peak_segment": "seg-06",
  "schema": "schema/dataset.schema.md",
  "schema_json": "schema/dataset.schema.json",
  "schema_version": "1.2.0",
  "segment_time_bands_dp": {
    "enabled": false
  },
  "summary": {
    "duplicates_removed": 1,
    "excluded_low_confidence_fraction": 0.0175,
    "exposure_coverage": 0.0667,
    "out_of_window": 0,
    "reports_in": 59,
    "segments_published": 177,
    "segments_total": 180,
    "segments_withheld_low_count": 3,
    "snapped": 57,
    "unsnapped": 1
  },
  "temporal": {
    "basis": "Report volume by time of day and day of week (city-wide aggregate). NOT an exposure-normalized rate: it reflects when people report, which tracks when they ride. Treat as a lead, not a risk ranking.",
    "by_hour": {
      "0": 3,
      "1": 3,
      "10": 3,
      "11": 3,
      "12": 3,
      "13": 3,
      "14": 3,
      "15": 3,
      "16": 3,
      "17": 3,
      "18": 3,
      "19": 3,
      "2": 2,
      "20": 3,
      "21": 3,
      "22": 3,
      "23": 3,
      "7": 2,
      "8": 3,
      "9": 3
    },
    "by_part_of_day": {
      "am_peak": 8,
      "evening": 12,
      "midday": 18,
      "overnight": 8,
      "pm_peak": 12
    },
    "by_weekday": {
      "Thu": 8,
      "Wed": 50
    },
    "peak_part_of_day": "midday",
    "peak_weekday": "Wed",
    "small_sample_caveat": false,
    "suppressed_low_count": false,
    "total_reports_timed": 58
  },
  "version": "0.1.2",
  "window": {
    "end": "2026-12-31",
    "start": "2026-01-01"
  }
}